Qualification of S-N Curve by Experiments: Estimation of Survival Probability for a Low Number of Experiments

Abstract

Due to the cost of fatigue testing, qualification of new weld details or improved welding techniques is often performed by comparing the experimental results with a mean S-N curve. The experimental dataset is considered to qualify for a given S-N curve if the mean of the log of the experimental results is larger than the mean of the chosen S-N curve plus an interval depending on the chosen confidence level. The confidence level is generally chosen arbitrary high, typically 75% or 90%. The survival estimation does not take into account the risk incurred by the fact that the actual mean has a probability of 25% or 10% of being lower than the theoretical S-N curve. In the present paper, we propose a different methodology based on the design S-N curve. The probability of failure for a given load level is computed by combining the probability of failure considering the mean value of the S-N curve, and the probability of this mean value. Thanks to the present method it is possible to estimate the probability of failure without making an assumption on the confidence level, which reduces the arbitrary during the estimation of the test results, particularly when the number of experimental results is limited.